Stage 1: the ticket meets a machine
RemediateDesk connects to Autotask and checks for updated tickets every couple of minutes. A ticket only starts something when two things are true: its device has been mapped to a managed endpoint, and a saved Autopilot matches the problem it describes.
That second condition is deliberate. A ticket with no matching Autopilot is not handed to a free-running AI in the background, because nobody would be there to answer its questions. It stays in the queue for a technician, who can open that machine's chat and work it with the AI beside them.
Stage 2: the work
The matched Autopilot is a set of written instructions. The AI follows them on the real machine, running one step, reading the output, and choosing the next. Each Autopilot carries a trust level that decides how much freedom it has:
- Assisted. Where every new Autopilot starts. Steps go through the normal approval gate, so a person sees what is about to run.
- Trusted. Reached after a run of verified successes. Still approval-gated, with a track record behind it.
- Zero-touch. Runs without an approval click. Your never-allowed list still applies, checked against the exact commands once any values are filled in, and where a step truly needs a person you write that into the instructions and it waits.
A person reads and saves the instructions before they ever run, and for Zero-touch that review is the approval. Levels are earned through verified runs, and a failed run demotes the Autopilot one level automatically. An administrator can also set a level by hand.
Stage 3: proving it worked
A fix that exits with code 0 has only proved the command was accepted. The platform requires a second, separate check after any change before it will mark a session resolved, and refuses to resolve if that check shows the problem is still there.
Four outcomes exist, and they are counted separately so the headline numbers cannot be flattered:
| Outcome | Meaning |
|---|---|
| Resolved | A change was made and a follow-up check confirmed it |
| No action needed | The alert was stale or a false alarm; nothing was wrong to fix |
| Insufficient evidence | It looked and could not safely conclude; a person is told what to try |
| Escalated | It reached its attempt limit and stopped |
Stage 4: what your PSA receives
A note is posted to the ticket. It is internal by default, so a customer never sees AI-written text unreviewed. For a hand-off, the note says what was checked and what a technician should do next, so whoever picks the ticket up starts from the findings, not from zero.
Close-out is a button, not a surprise: the AI drafts the summary and a technician posts it and marks the ticket complete. The platform then reads the ticket back from the PSA to confirm it really closed.
What this is not
It is not a replacement for the person who owns the queue. It removes the first twenty minutes of repetitive investigation from the tickets it can handle, and tells you plainly about the ones it cannot. See the real session for what that looks like with timestamps.